P-Value Segment Selection Technique for Speaker Verification

Abstract

This paper presents a segment selection technique for discarding portions of speech that result in poor discrimination ability in speaker verification tasks. Theory supporting the significance of a frame selection procedure for test segments, prior to making decisions, is also developed. This approach has the ability to reduce the effect of the acoustic regions of speech that are not accurately represented due to a lack of training data. Compared with a baseline system using both CMS and variance normalization, the proposed segment selection technique brings 24% relative reduction in error rate over the entire testing data of the 2002 NIST Dataset in terms of minimum DCF. For short test segments, i.e. less than 15 seconds, the novel frame dropping technique produces a significant relative error rate reduction of 23% in terms of minimum DCF.

Constrained temporal structure for text-dependent speaker verification

“…However such systems, which are invariably founded on the GMM/UBM paradigm [2], exhibit high sensitivity to the quantity of data, particularly the reference model data [3], [4], [5]. Their performance degrades strongly while reducing the duration of speech material available [6,7,8]. For situations where the speech duration is below 30 seconds, recognition performance falls rapidly [9,10].…”

Score weighting in speaker verification systems

“…Thus phonetic content variation, in addition to other factors such as the variability of the feature vector distribution from session to session, and MAP adaptation itself, has made some regions of the feature space less reliable in making the final decision. We have addressed the score variability caused by the lack of training data in our previous work [6,7] by dropping the nondiscriminative frames according to their target and impostor scores without making any a priori assumptions about the distributions of impostor and target scores.…”

“…Following on from our previous investigations [6,7], we now address the score variability caused by phonetic variation by emphasising the best scoring GMM frames that are strongly correlated with particular phonemes e.g. vowels and nasals [3].…”